A method for predicting lighting equipment failures, a terminal device, and a storage medium.

Abstract

This invention relates to a method, terminal device, and storage medium for predicting lighting equipment faults. The method includes: collecting state parameters of the lighting equipment over a period of time; constructing a training set based on the collected state parameters; constructing a lighting equipment state parameter prediction model based on an improved LSTM and attention mechanism, introducing a memory gate in the traditional LSTM to distinguish between short-term and long-term memory updates, allowing the model to employ different strategies when processing short-term and long-term memory; predicting the state parameters of the lighting equipment using the trained model; performing a health score on the lighting equipment based on the predicted state parameters; and predicting the faults of the lighting equipment based on the health score results. This invention improves the real-time performance and accuracy of lighting equipment fault prediction and reduces the probability of fault occurrence.

Description

Technical Field

[0001] This invention relates to the field of equipment failure prediction, and more particularly to a method for predicting failures in lighting equipment, a terminal device, and a storage medium. Background Technology

[0002] As a crucial component of urban infrastructure, the stable operation of urban lighting systems directly impacts urban safety and energy efficiency. However, traditional streetlight maintenance methods are largely reactive, addressing issues only after they occur, leading to untimely troubleshooting and high maintenance costs. Furthermore, while existing IoT applications have achieved basic equipment status monitoring, they lack efficient predictive mechanisms, failing to accurately anticipate potential equipment failures and hindering proactive maintenance. Summary of the Invention

[0003] To address the aforementioned problems, this invention proposes a method for predicting lighting equipment faults, a terminal device, and a storage medium.

[0004] The specific plan is as follows:

[0005] A method for predicting lighting equipment failures includes the following steps:

[0006] S1: Collect the state parameters of the lighting equipment over a period of time, and build a training set based on the collected state parameters;

[0007] S2: Construct a lighting device state parameter prediction model based on an improved LSTM and attention mechanism, and train the model using a training set;

[0008] The improved LSTM is based on the traditional LSTM and includes the following improvements:

[0009] (1) Change the method of using the same weight matrix for the hidden state at the previous time step and the input at the current time step in the forget gate, input gate and output gate to use different weight matrices;

[0010] (2) Introducing a memory gate allows the model to selectively retain or discard short-term and long-term memories; the memory gate is represented as:

[0011] g t =σ(W g ·x t +U g ·h t-1 +b g )

[0012] Among them, g t Indicates the output of the memory gate at the current moment; x t Indicates the input at the current time; W g The weight matrix representing the effect of the current input on the memory gate; ht-1 Indicates the hidden state at the previous moment; U g The weight matrix representing the influence of the hidden state at the previous time step on the memory gate; b g For the bias term in the memory gate;

[0013] (3) Adjust the update method of the cell state using the following formula:

[0014]

[0015] Among them, C t and C t-1 These represent the cell states at the current and previous times, respectively; f t Indicates the output of the forget gate at the current moment; i t This indicates the input gate output at the current moment; Represents the current state of candidate memory cells; ⊙ represents the Hadamard product; tanh represents the hyperbolic tangent function; W c The weight matrix representing the influence of the current input on the candidate memory; U c The weight matrix representing the influence of the previous hidden state on the candidate memory; b c For the bias term in the update of candidate memory cell state;

[0016] S3: Predict the state parameters of lighting equipment using the trained model;

[0017] S4: Perform health scoring on lighting equipment based on predicted state parameters;

[0018] S5: Predict lighting equipment failures based on health score results.

[0019] Furthermore, the status parameters include electrical parameters, physical status parameters, and environmental status parameters; electrical parameters include voltage, current, power, and three-phase voltage; physical status parameters include lamp body temperature and equipment life statistics; and environmental status parameters include ambient humidity.

[0020] Furthermore, the formula for scoring the health of lighting equipment is as follows:

[0021]

[0022] Among them, H i P represents the health score of the i-th lighting device; jmax w represents the maximum allowed value of the j-th state parameter; j y represents the weight of the j-th state parameter; j P represents the predicted value of the j-th state parameter; j represents the index of the state parameter; n represents the total number of state parameters; jmax This represents the maximum allowed value for the j-th state parameter.

[0023] Furthermore, the method for predicting lighting equipment failures based on health score results is as follows: determine whether the health score is less than a score threshold; if so, perform failure prediction and calculate the probability of failure occurrence using the following formula:

[0024]

[0025] Where, ΔX t α represents the difference between the predicted value of the state parameter and the baseline value; α, β, and γ all represent model parameters, which are obtained by training with historical data; Δt represents the time difference between the prediction time and the current time.

[0026] Furthermore, it also includes performing fault maintenance when the calculated probability of a fault occurring is greater than a probability threshold.

[0027] A lighting equipment fault prediction terminal device includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the method described above in the embodiments of the present invention.

[0028] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described above in the embodiments of the present invention.

[0029] The present invention adopts the above technical solution, which improves the real-time performance and accuracy of lighting equipment fault prediction and reduces the probability of fault occurrence. Attached Figure Description

[0030] Figure 1 The diagram shown is a flowchart of a method according to an embodiment of the present invention. Detailed Implementation

[0031] To further illustrate the various embodiments, the present invention provides accompanying drawings. These drawings are part of the disclosure of the present invention, primarily used to illustrate the embodiments, and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementations and the advantages of the present invention.

[0032] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments.

[0033] Example 1:

[0034] This invention provides a method for predicting lighting equipment faults, such as... Figure 1 As shown, the method includes the following steps:

[0035] S1: Collect the state parameters of lighting equipment (such as streetlights) over a period of time, and build a training set based on the collected state parameters.

[0036] Comprehensive data sources can enhance the dimensionality of the model's input features, thereby improving prediction accuracy. In this embodiment, electrical parameters, physical state parameters, and environmental state parameters are all used as state parameters for subsequent model training to improve the accuracy of health scores and generate more reasonable maintenance plans.

[0037] The status parameters of lighting equipment are collected in real time through IoT terminal devices, including:

[0038] Electrical parameters: voltage, current, power, three-phase voltage;

[0039] Physical condition parameters: lamp body temperature, equipment life statistics;

[0040] Environmental condition parameters: ambient humidity.

[0041] All data is stored in time-series format and preprocessed to convert it into a standard format for training LSTM models.

[0042] S2: Construct a lighting device state parameter prediction model based on an improved LSTM and attention mechanism, and train the model using a training set.

[0043] Traditional LSTM models can suffer from information confusion when processing long-term and short-term memory. In LSTM models, traditional memory management strategies typically control the flow of information through forget gates, input gates, and output gates. However, these gating mechanisms are usually fixed and cannot dynamically adjust their ability to process short-term and long-term memory. Especially when faced with complex time-series data, the model may fail to effectively distinguish which information should be stored long-term and which should be stored short-term, leading to confusion in the "memory" of certain important information in the time series. The goal of adaptive memory management strategies is to enable the model to dynamically adjust its processing of short-term and long-term memory based on the characteristics of the input data, the context, and feedback during the learning process. This allows LSTM to more accurately capture short-term changes and long-term trends in time series data.

[0044] This embodiment employs a separation mechanism, introducing a dynamic gating mechanism (memory gate) to distinguish between short-term and long-term memory updates, allowing the model to use different strategies when processing short-term and long-term memories. This improves the model's ability to capture instantaneous changes and long-term trends in time series data.

[0045] LSTM is used to capture long-term dependencies in device states and predict future trends of input state parameters. The hidden state h at each time step... tThis represents the state characteristics of the lighting equipment at that moment. LSTM uses a gating mechanism to capture long-term dependencies in the time series. The input gate controls the importance of each state parameter (such as temperature and voltage) in the prediction; the forget gate discards irrelevant historical states.

[0046] 1. Forgotten Gate

[0047] Decide how much of the state information from the previous time step needs to be forgotten.

[0048] f t =σ(W f ·x t +U f ·h t-1 +b f )

[0049] Among them, f t The current forget gate output controls how much of the past memory the model needs to forget during updates. σ represents the sigmoid activation function, which limits the output to between 0 and 1, representing the degree of forgetting (0 for complete forgetting, 1 for complete retention). W f ·x t This indicates the current input x. t (i.e., the current state parameters) and the weight matrix W f The product of and represents the effect of the current input on the forget gate. U f ·h t-1 Indicates the hidden state h in the previous moment. t-1 With weight matrix U f The product of and represents the effect of the previous moment's memory on the forgetting gate. f It is a bias term that helps the model output meaningful values ​​even when there is no input or hidden state, thus enhancing the model's flexibility.

[0050] 2. Input Gate

[0051] It determines the impact of the current input information on the cell state.

[0052] i t =σ(W i ·x t +U i ·h t-1 +b i )

[0053] Among them, i t The current input gate output controls the current input x. t The degree to which the state of the memory cell is updated. σ is the sigmoid activation function, ensuring the output is between 0 and 1, representing the degree to which input information should be retained. W i ·x tIndicates the current input x t With weight matrix W i The product of and represents the effect of the current input on the input gate. i ·h t-1 The hidden state h represents the state at the previous time step. t-1 With weight matrix U i The product of and represents the influence of the previous time step's memory on the current input gate. i It is a bias term that increases the flexibility of the input gate and helps the model adjust better based on historical data.

[0054] 3. Output gate

[0055] Determines the output of the hidden state at the current time step.

[0056] o t =σ(W o ·x t +U o ·h t-1 +b o )

[0057] Among them, o t The output gate at the current moment represents the output value, which controls the final output of the model and determines the update of the hidden state. σ is the sigmoid activation function, ensuring the output is between 0 and 1, indicating how much memory to output. W o ·x t This indicates the effect of the current input on the output gate. U o ·h t-1 This indicates the effect of the previous hidden state on the current output gate. o It is a bias term that adjusts the flexibility of the output gate.

[0058] In a traditional LSTM, all update steps (such as forgetting, input, and output) are based on the current input x. t and the hidden state h from the previous moment t-1 The decision is made together. This is achieved by introducing a weight matrix U(U f U i U o This allows the current state update to be influenced not only by the current input but also by the past state h. t-1 This influences the model, allowing it to have a stronger historical dependency when updating.

[0059] For example, we want to predict how the temperature of streetlight equipment changes over time. The input at the current moment might be information such as the current temperature, ambient humidity, and power (which will be determined by x). t (represented by h), while the equipment status and temperature change trend of the previous moment may affect the temperature prediction at the current moment (this is determined by h). t-1express).

[0060] The weight matrix W is responsible for weighting the current input information such as temperature, humidity, and power before feeding it into the model, thus affecting the current state update.

[0061] Weight matrix U(U f U i U o ): Responsible for feeding the weighted equipment state information (such as temperature and power in the previous moment) into the model, so that the model can better capture the time dependence of the equipment state.

[0062] 4. Memory gate (an additional gating mechanism)

[0063] An additional parameter, weight Wg, controls the impact of input features on short-term memory. It determines the degree to which short-term memory contributes to the final state at the current time step.

[0064] g t =σ(W g ·x t +U g ·h t-1 +b g )

[0065] Among them, g t This represents the output of the memory gate at the current moment, controlling the introduction of short-term memory; W g The weight matrix representing the effect of the current input on the memory gate; U g The weight matrix representing the influence of the hidden state at the previous time step on the memory gate; b g It is a bias term, which further enhances the adjustment capability of the memory gate.

[0066] 5. Candidate memory unit update

[0067]

[0068] in, The current candidate memory state represents a new memory generated from the current input and the hidden state from the previous time step. tanh is the hyperbolic tangent activation function, which outputs memory states between -1 and 1, helping to normalize the data. W c This represents the weight matrix indicating the influence of the current input on the candidate memory. c b is a weight matrix representing the influence of the previous hidden state on the candidate memory. c It is a bias term that helps adjust the output of candidate memories.

[0069] 6. Cell status update

[0070]

[0071] Among them, C t Indicates the current cell state; C t-1 This represents the cell state at the previous time step, storing information for long-term memory; ⊙ is the Hadamard product (element-by-element multiplication), ensuring that the fusion of each part contributes reasonably to the final state; f t ⊙C t-1 The forget gate indicates the degree to which memories from the previous moment are retained; The input gate, candidate memory, and memory gate work together to determine how the current input affects the current cell state.

[0072] 7. Hidden state update

[0073] Calculate the hidden state of the current time step based on the cell state and the output gate.

[0074] h t =o t tanh(C t )

[0075] Among them, h t This represents the hidden state at the current time step, and the final output is the model's state, used for prediction and calculation at the next time step; t The output gate controls the degree of output of the hidden state at the current moment; tanh(C t The current state of the cell is represented by the hyperbolic tangent function, which is used to normalize the output to ensure that it is within a reasonable range.

[0076] The attention mechanism introduced in this embodiment is used to make the model pay more attention to important time steps, as follows:

[0077] 1. Calculate the importance score for each time step t:

[0078] e t =v T ·tanh(W s ·h t )

[0079] Among them, W s represents the training parameter matrix of the model, used for feature transformation; v represents a vector in the model, used for score aggregation.

[0080] Importance score e t This represents the potential contribution of the data to future predictions at time step t. During the scoring process, the hidden state h at each time step... t This formula quantifies it into a fraction e. t And further used to calculate the attention weight α at that time step. t .

[0081] 2. Based on importance score e t Calculate the attention weight α at each time step. t :

[0082]

[0083] Attention weight α t This determines the degree of influence of each time step on the prediction result.

[0084] Note: In some cases, the weight α can be used directly. t This is insufficient to fully reflect the importance of the data. Therefore, the system can use an importance score e. t Come:

[0085] (1) Filtering key time steps:

[0086] When e at certain time steps t When the data exceeds a set threshold, these time steps can be marked as high priority for further analysis.

[0087] (2) Dynamically adjust the input sequence:

[0088] When dealing with large amounts of data (such as generating device status data once per minute), the system can eliminate low-priority time steps and retain the most critical parts for the prediction results, thereby reducing the computational burden and improving efficiency.

[0089] 3. Weight-based prediction results

[0090] The final prediction is a weighted sum of all time steps, ensuring that the system focuses on the time step data that is most valuable for the prediction.

[0091]

[0092] Suppose we predict the device state parameters for several future time steps, and the original prediction result is as follows:

[0093]

[0094] The prediction result after adding attention weights is:

[0095]

[0096] S3: Predict the state parameters of lighting equipment using the trained model.

[0097] S4: Perform a health score on lighting equipment based on predicted state parameters.

[0098] In this embodiment, the health score is set using the following formula:

[0099]

[0100] Among them, H i y represents the health score of the i-th device; j P represents the predicted value of the j-th state parameter; j represents the index of the state parameter; n represents the total number of state parameters; jmax This represents the maximum permissible value of the j-th state parameter (i.e., the upper limit of this state parameter under normal conditions, such as the maximum permissible value of voltage being 240V, exceeding this value may indicate an abnormality); w j This represents the weight of the j-th state parameter, which can be set by those skilled in the art according to their needs.

[0101] S5: Predict lighting equipment failures based on health score results.

[0102] The method for predicting lighting equipment failures based on health scores is as follows: determine whether the health score is less than a threshold (e.g., 50 points). If so, perform a failure prediction and calculate the probability P of the failure using the following formula. Failure :

[0103]

[0104] Where, ΔX t α represents the difference between the predicted value of the state parameter and the baseline value (determined by the average value collected when no fault occurs); α, β, and γ all represent model parameters, which are trained from historical data; Δt represents the time difference between the prediction time (i.e., the prediction time corresponding to the predicted value of the state parameter) and the current time.

[0105] Furthermore, this embodiment also includes performing fault maintenance when the predicted probability of a fault occurrence is greater than a set probability threshold. The following strategy is adopted in this embodiment:

[0106] (1) Maintenance type: preventive inspection, component replacement, etc.;

[0107] (2) Priority: Automatically sorted according to failure probability;

[0108] (3) Resource allocation: Dispatch available maintenance personnel and equipment parts.

[0109] For example, if the risk of voltage anomalies and temperature rises is high, the system recommends scheduling equipment inspections within one week and preparing new heat dissipation modules in advance. In practice, those skilled in the art can develop corresponding fault maintenance strategies based on specific applications; no restrictions are imposed here.

[0110] Example 2:

[0111] The present invention also provides a lighting equipment fault prediction terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the method embodiment described above in Embodiment 1 of the present invention.

[0112] Furthermore, as an executable solution, the lighting equipment fault prediction terminal device can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. The lighting equipment fault prediction terminal device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above-described composition of the lighting equipment fault prediction terminal device is merely an example and does not constitute a limitation on the lighting equipment fault prediction terminal device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the lighting equipment fault prediction terminal device may also include input / output devices, network access devices, buses, etc., and this embodiment of the invention does not limit this.

[0113] Furthermore, as an executable solution, the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices. The general-purpose processor can be a microprocessor or any conventional processor. This processor serves as the control center of the lighting equipment fault prediction terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0114] The memory can be used to store the computer programs and / or modules. The processor, by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory, realizes various functions of the lighting equipment fault prediction terminal device. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0115] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method described in the embodiments of the present invention.

[0116] If the modules / units integrated in the lighting equipment fault prediction terminal device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), and software distribution media, etc.

[0117] Although the invention has been specifically shown and described in conjunction with preferred embodiments, those skilled in the art should understand that various changes in form and detail may be made to the invention without departing from the spirit and scope of the invention as defined in the appended claims, all of which shall be within the scope of protection of the invention.

Claims

1. A method for predicting faults in lighting equipment, characterized in that, Includes the following steps: S1: Collect the state parameters of the lighting equipment over a period of time, and build a training set based on the collected state parameters; S2: Construct a lighting device state parameter prediction model based on an improved LSTM and attention mechanism, and train the model using a training set; The improved LSTM is based on the traditional LSTM and includes the following improvements: (1) Change the method of using the same weight matrix for the hidden state at the previous time step and the input at the current time step in the forget gate, input gate and output gate to use different weight matrices; (2) Introducing a memory gate allows the model to selectively retain or discard short-term and long-term memories; the memory gate is represented as: in, This represents the output of the memory gate at the current moment; This indicates the input at the current time. The weight matrix represents the influence of the current input on the memory gate; This indicates the previously hidden state; The weight matrix represents the influence of the hidden state in the previous time step on the memory gate; For the bias term in the memory gate; (3) Adjust the update method of the cell state using the following formula: in, and These represent the cell states at the current and previous times, respectively. This indicates the output of the forget gate at the current moment; This indicates the input gate output at the current moment; Indicates the current state of candidate memory cells; ⊙ represents the Hadamard product; Represents the hyperbolic tangent function; The weight matrix represents the influence of the current input on the candidate memory. The weight matrix represents the influence of the hidden state at the previous time step on the candidate memory; For the bias term in the update of candidate memory cell state; S3: Predict the state parameters of lighting equipment using the trained model; S4: Perform health scoring on lighting equipment based on predicted state parameters; S5: Predict lighting equipment failures based on health score results.

2. The lighting equipment fault prediction method according to claim 1, characterized in that: The status parameters include electrical parameters, physical status parameters, and environmental status parameters; electrical parameters include voltage, current, power, and three-phase voltage; physical status parameters include lamp body temperature and equipment life statistics; environmental status parameters include ambient humidity.

3. The lighting equipment fault prediction method according to claim 1, characterized in that: The formula for scoring the health of lighting equipment is: in, This represents the health score of the i-th lighting device; This represents the maximum allowed value of the j-th state parameter; This represents the weight of the j-th state parameter; This represents the predicted value of the j-th state parameter; j represents the index of the state parameter; and n represents the total number of state parameters.

4. The lighting equipment fault prediction method according to claim 3, characterized in that: The method for predicting lighting equipment failures based on health scores is as follows: determine whether the health score is less than a score threshold; if so, perform a failure prediction by calculating the probability of failure using the following formula: in, This represents the difference between the predicted value of the state parameter and the baseline value. , , All of these represent model parameters, obtained through training with historical data; This represents the time difference between the predicted time and the current time.

5. The lighting equipment fault prediction method according to claim 4, characterized in that: It also includes performing fault maintenance when the calculated probability of a fault occurring is greater than a probability threshold.

6. A terminal device for predicting lighting equipment faults, characterized in that: It includes a processor, a memory, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the method as described in any one of claims 1 to 5.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 5.

Images